Message-ID: <8ddff47d-3059-42ea-b022-6151da513049@lucifer.local>
Date: Wed, 6 Aug 2025 20:02:59 +0100
From: Lorenzo Stoakes <lorenzo.stoakes@...cle.com>
To: Suren Baghdasaryan <surenb@...gle.com>
Cc: akpm@...ux-foundation.org, Liam.Howlett@...cle.com, david@...hat.com,
        vbabka@...e.cz, peterx@...hat.com, jannh@...gle.com,
        hannes@...xchg.org, mhocko@...nel.org, paulmck@...nel.org,
        shuah@...nel.org, adobriyan@...il.com, brauner@...nel.org,
        josef@...icpanda.com, yebin10@...wei.com, linux@...ssschuh.net,
        willy@...radead.org, osalvador@...e.de, andrii@...nel.org,
        ryan.roberts@....com, christophe.leroy@...roup.eu,
        tjmercier@...gle.com, kaleshsingh@...gle.com, aha310510@...il.com,
        linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
        linux-mm@...ck.org, linux-kselftest@...r.kernel.org
Subject: Re: [PATCH v3 3/3] fs/proc/task_mmu: execute PROCMAP_QUERY ioctl
 under per-vma locks

On Wed, Aug 06, 2025 at 08:59:04AM -0700, Suren Baghdasaryan wrote:
> Utilize per-vma locks to stabilize vma after lookup without taking
> mmap_lock during PROCMAP_QUERY ioctl execution. If vma lock is
> contended, we fall back to mmap_lock but take it only momentarily
> to lock the vma and release the mmap_lock. In a very unlikely case
> of vm_refcnt overflow, this fall back path will fail and ioctl is
> done under mmap_lock protection.
>
> This change is designed to reduce mmap_lock contention and prevent
> PROCMAP_QUERY ioctl calls from blocking address space updates.
>
> Signed-off-by: Suren Baghdasaryan <surenb@...gle.com>

A lot of nits but nothing's really standing out as broken, AFAICT...

> ---
>  fs/proc/task_mmu.c | 84 +++++++++++++++++++++++++++++++++++++---------
>  1 file changed, 68 insertions(+), 16 deletions(-)
>
> diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> index 45134335e086..0396315dfaee 100644
> --- a/fs/proc/task_mmu.c
> +++ b/fs/proc/task_mmu.c
> @@ -517,28 +517,81 @@ static int pid_maps_open(struct inode *inode, struct file *file)
>  		PROCMAP_QUERY_VMA_FLAGS				\
>  )
>
> -static int query_vma_setup(struct mm_struct *mm)
> +#ifdef CONFIG_PER_VMA_LOCK
> +
> +static int query_vma_setup(struct proc_maps_locking_ctx *lock_ctx)
>  {
> -	return mmap_read_lock_killable(mm);
> +	lock_ctx->locked_vma = NULL;
> +	lock_ctx->mmap_locked = false;

We also do this in lock_vma_range(), seems sensible to factor out? E.g.:

static void ctx_clear_vma(struct proc_maps_locking_ctx *lock_ctx)
{
	lock_ctx->locked_vma = NULL;
	lock_ctx->mmap_locked = false;
}

> +
> +	return 0;
>  }
>
> -static void query_vma_teardown(struct mm_struct *mm, struct vm_area_struct *vma)
> +static void query_vma_teardown(struct proc_maps_locking_ctx *lock_ctx)
>  {
> -	mmap_read_unlock(mm);
> +	if (lock_ctx->mmap_locked)
> +		mmap_read_unlock(lock_ctx->mm);

Maybe worth a comment as to why we leave lock_ctx->mmap_locked set here?

> +	else
> +		unlock_vma(lock_ctx);

Should have said on 2/3, but I wonder if we should prefix with ctx_, as
'unlock_vma()' and 'lock_vma()' seem like core functions... esp. since we
have vma_start_read/write() rather than functions that reference locking.

So - ctx_unlock_vma() and ctx_lock_vma() or unlock_ctx_vma() /
lock_ctx_vma()?

> +}
> +
> +static struct vm_area_struct *query_vma_find_by_addr(struct proc_maps_locking_ctx *lock_ctx,
> +						     unsigned long addr)
> +{
> +	struct vm_area_struct *vma;
> +	struct vma_iterator vmi;
> +
> +	if (lock_ctx->mmap_locked)
> +		return find_vma(lock_ctx->mm, addr);
> +
> +	unlock_vma(lock_ctx);
> +	rcu_read_lock();
> +	vma_iter_init(&vmi, lock_ctx->mm, addr);
> +	vma = lock_next_vma(lock_ctx->mm, &vmi, addr);
> +	rcu_read_unlock();

I think a comment at the top of this block would be useful, something like
'We unlock any previously locked VMA and find the next under RCU'.

> +
> +	if (!IS_ERR_OR_NULL(vma)) {

Is the NULL bit here really necessary? presumably lock_ctx->locked_vma is
expected to be NULL already, so we're not overwriting anything here.

In fact we could get rid of the horrid if/else here with a guard clause like:

	if (!IS_ERR(vma) || PTR_ERR(vma) != -EAGAIN)
		return vma;

(the !IS_ERR() bit is probably a bit redundant but makes things clearer
vs. just the PTR_ERR() thing)

Then do the rest below.


> +		lock_ctx->locked_vma = vma;
> +	} else if (PTR_ERR(vma) == -EAGAIN) {
> +		/* Fallback to mmap_lock on vma->vm_refcnt overflow */
> +		mmap_read_lock(lock_ctx->mm);
> +		vma = find_vma(lock_ctx->mm, addr);
> +		lock_ctx->mmap_locked = true;

Kinda sucks we have two separate ways of doing fallback now, this
open-coded thing and fallback_to_mmap_lock().

Sort of hard to combine since we have subtly diffrent semantics - the RCU
read lock is being held in the /proc/$pid/maps case, but here we've
released it already.

> +	}
> +
> +	return vma;
> +}
> +
> +#else /* CONFIG_PER_VMA_LOCK */
> +
> +static int query_vma_setup(struct proc_maps_locking_ctx *lock_ctx)
> +{
> +	return mmap_read_lock_killable(lock_ctx->mm);
>  }
>
> -static struct vm_area_struct *query_vma_find_by_addr(struct mm_struct *mm, unsigned long addr)
> +static void query_vma_teardown(struct proc_maps_locking_ctx *lock_ctx)
>  {
> -	return find_vma(mm, addr);
> +	mmap_read_unlock(lock_ctx->mm);
>  }
>
> -static struct vm_area_struct *query_matching_vma(struct mm_struct *mm,
> +static struct vm_area_struct *query_vma_find_by_addr(struct proc_maps_locking_ctx *lock_ctx,
> +						     unsigned long addr)
> +{
> +	return find_vma(lock_ctx->mm, addr);
> +}
> +
> +#endif  /* CONFIG_PER_VMA_LOCK */
> +
> +static struct vm_area_struct *query_matching_vma(struct proc_maps_locking_ctx *lock_ctx,
>  						 unsigned long addr, u32 flags)
>  {
>  	struct vm_area_struct *vma;
>
>  next_vma:
> -	vma = query_vma_find_by_addr(mm, addr);
> +	vma = query_vma_find_by_addr(lock_ctx, addr);
> +	if (IS_ERR(vma))
> +		return vma;
> +
>  	if (!vma)
>  		goto no_vma;
>
> @@ -579,11 +632,11 @@ static struct vm_area_struct *query_matching_vma(struct mm_struct *mm,
>  	return ERR_PTR(-ENOENT);
>  }
>
> -static int do_procmap_query(struct proc_maps_private *priv, void __user *uarg)
> +static int do_procmap_query(struct mm_struct *mm, void __user *uarg)
>  {
> +	struct proc_maps_locking_ctx lock_ctx = { .mm = mm };

>  	struct procmap_query karg;
>  	struct vm_area_struct *vma;
> -	struct mm_struct *mm;
>  	const char *name = NULL;
>  	char build_id_buf[BUILD_ID_SIZE_MAX], *name_buf = NULL;
>  	__u64 usize;
> @@ -610,17 +663,16 @@ static int do_procmap_query(struct proc_maps_private *priv, void __user *uarg)
>  	if (!!karg.build_id_size != !!karg.build_id_addr)
>  		return -EINVAL;
>
> -	mm = priv->lock_ctx.mm;
>  	if (!mm || !mmget_not_zero(mm))
>  		return -ESRCH;
>
> -	err = query_vma_setup(mm);
> +	err = query_vma_setup(&lock_ctx);
>  	if (err) {
>  		mmput(mm);
>  		return err;
>  	}
>
> -	vma = query_matching_vma(mm, karg.query_addr, karg.query_flags);
> +	vma = query_matching_vma(&lock_ctx, karg.query_addr, karg.query_flags);
>  	if (IS_ERR(vma)) {
>  		err = PTR_ERR(vma);
>  		vma = NULL;
> @@ -705,7 +757,7 @@ static int do_procmap_query(struct proc_maps_private *priv, void __user *uarg)
>  	}
>
>  	/* unlock vma or mmap_lock, and put mm_struct before copying data to user */
> -	query_vma_teardown(mm, vma);
> +	query_vma_teardown(&lock_ctx);
>  	mmput(mm);
>
>  	if (karg.vma_name_size && copy_to_user(u64_to_user_ptr(karg.vma_name_addr),
> @@ -725,7 +777,7 @@ static int do_procmap_query(struct proc_maps_private *priv, void __user *uarg)
>  	return 0;
>
>  out:
> -	query_vma_teardown(mm, vma);
> +	query_vma_teardown(&lock_ctx);
>  	mmput(mm);
>  	kfree(name_buf);
>  	return err;
> @@ -738,7 +790,7 @@ static long procfs_procmap_ioctl(struct file *file, unsigned int cmd, unsigned l
>
>  	switch (cmd) {
>  	case PROCMAP_QUERY:
> -		return do_procmap_query(priv, (void __user *)arg);
> +		return do_procmap_query(priv->lock_ctx.mm, (void __user *)arg);

OK this confused me until I worked it through.

We set priv->lock_ctx.mm in:

pid_maps_open() -> do_maps_open() -> proc_maps_open()

Which it gets from proc_mem_open() which figures out the mm.

Maybe one for 2/3, but it'd be nice to have a comment saying something
about how this is set, since it being part of lock_ctx makes it seem like
it's something that would be set elsewhere.

Since we have fallback stuff and want to thread through this new lokc
context type I guess it makes sense to put it here but given that's the
case, let's maybe just add a comment here to clarify.

>  	default:
>  		return -ENOIOCTLCMD;
>  	}
> --
> 2.50.1.565.gc32cd1483b-goog
>

Cheers, Lorenzo

Powered by blists - more mailing lists